In one of your earlier videos, I think you mentioned why AMOS is better than PLS based SEM software. Now your recent videos are about Smart PLS. So, I wonder if you have changed your mind about using AMOS? Are you going to use SmartPLS from now on instead of AMOS? Or are you planning to use AMOS and SmartPLS for different purposes?
AMOS is still definitely better for models with all reflective models, because it allows the estimation of model fit. SmartPLS is better for many other things, such as moderation, mediation, and complex things like segmentation (FIMIX and POS). So, I suppose I like AMOS for measurement models and PLS for structural models :) (except the lack of model fit for structural models...)
Thanks for the video Dr. I got different results in AVE, Cronbach's, and P values of the path when using the Process feature, Consistent PLS, and general PLS. Cronbach's are all over 0.7 when using cPLS and general PLS, while some are below 0.7 when using PROCESS (0.59 is the worst). AVEs are all above 0.5 when using general PLS while some are below 0.5 (the worst is 0.4) when using cPLS. I remember you said if my indicators are all reflective I should choose cPLS, but the P values of the paths in cPLS are too perfect (all 0.000), while when using general PLS they are more "realistic", one is 0.09 while all others are below 0.05... My model has three controlled variables but no moderation. I totally do not know which result to trust. Also, because many of my indicators are not normally distributed, I choose PLS instead of CB-SEM. Thank you for helping!
If all factors are reflective, it is probably best to use cPLS. I would not rely in PROCESS for anything except a causal model, which you can also just do with cPLS.
Thank you so much Dr. I have watched almost all of your videos and thank you so much for helping new researchers like me! Great videos and all of them are very clear! Many many thanks and wish you all the best in your life and research!@@Gaskination
Hi Dr. Gaskin I was looking for videos that discuss control variables in CB-SEM models but I couldn't find any. I have one question regarding my model. Can I assess the model fit without the control variables or must they be included when checking fit? Thanks!
In a structural model, model fit should be assessed with all variables present that will be present when hypothesis testing. In a measurement model, just include all latent variables.
@@Gaskination I have four control variables, that are all categorical in nature, each with four categories. As such, I have created dummy variables and included n-1 when running the model (in order to have a reference category). However, when including these control variables, do I group each CV as a single latent variable or do I include each dummy as its own independent variable? I have seen people do both but I can't seem to get a clear answer.
Thank you for the video! Does this mean that we can now use "PROCESS" for analyzing Control Variables? Also, don't we need to handle the Dummy Variables? Does SmartPLS 4 somehow automatically handle categorical variables? How about Likert-scale variables as controls? Your response would be highly appreciated and helpful to many. Thanks.
1. Process in SmartPLS can include control variables. 2. Dummy variables must still be created for categorical variables. 3. Likert scales can always be used as controls.
@@Gaskination Huge thanks for the prompt response, Sir! Am I wrong in thinking that Likert scale measured construct is sort of categorical (ordinal). And wouldn't that mean it's necessary to create Dummy variable for all Likert-based constructs?
@@juvandi There is a lot of precedent for using Likert scales intact. Some statistical programs, like Mplus, do label ordinal variables as categorical, but they can still be used like a continuous variable. This is not true of nominal variables like job category or marital status, which need to be split into dummy variables. The difference is that with ordinal variables, an increase in the scale value equates to an increase in the construct trait value, whereas with nominal variables, an increase in the scale value does not equate to an increase in the trait value. For example, if with marital status, 1=single, 2=married, 3=divorced, 4=separated, 5=widowed, 6=living together unmarried, then a 5 on the scale is not "more marital status" than a 2. That just doesn't make any sense. Whereas, for a Likert scale for something like job satisfaction, a 5 ("strongly agree") does imply "more job satisfaction" than a 1 on that scale ("strongly disagree"). Hope this helps.
The SmartPLS website and books strongly discourage the assessment of model fit in PLS models. They have only reluctantly included any model fit features. I didn't make a video for Gaussian Copula, since it is still somewhat under debate whether it is broadly applicable or too sensitive for all but a few edge conditions. I'm not sure what NCA is.
Hello! Hope you are doing great. Actually I need guidance regarding usage of smart pls process feature. If we select process feature of smart pls 4 for structural model assessment, what we will do for measurement model assessment? Will we combine both the techniques of smart pls 4 i.e. pls sem for assessment of measurement model and process for structural model?? Kindly guide as I couldn't find any suitable answer from literature.
Yes, you can assess the measurement model in CB-SEM (if all reflective and good sample size) or PLS-SEM if it includes small sample size or formative factors.
Dear Dr. Gaskin, great and helpful video - again! But I have a question regarding the moderators in PROCESS, and it would be helpful if you could advice me.. In path coefficients, any moderation is not significant. In conditional direct effects, however, some are significant. As an example, I would like to test the hypothesis that A (IV) on B(DV) is moderated by age so that its effect is stronger for older people. If A -> B conditioned on age is significant at +1 SD and its Pathcoefficients are also greater than both, -1SD and at the mean, am I allowed to accept my hypothesis based on the specified conditions? (Even if in path coefficients AxAge→ B is not significant? ) I would be more than grateful if you could help me Kind regards, Hannah
There are many approaches to testing moderated effects. The most rigorous is probably floodlight analysis (not available in SmartPLS). The next is probably the simple-slopes analysis (+/- 1 std deviation). The least valid is probably looking directly at the coefficient of the interaction effect.
I cannot find a way to apply a quadradic effect to a moderation in SmartPLS 4. Perhaps a workaround would be to calculate an interaction term outside of SmartPLS, and then add a quadradic effect to the line that goes from the interaction to the DV.
@@Gaskination these measurement properties should be checked through cfa or measurement model of pls sem?? If we are going to check structural hypothesis through process feature of smart pls 4?
Pardon me Dr. Gaskin, may I know what is differences between conditional direct effects and conditional indirect effects? Could you interpret it? thank you so much :)
The word "conditional" just refers to moderation, and "indirect" refers to mediation. So, a conditional direct effect is just a moderated effect. A conditional indirect effect is the moderated mediated effect.
@@mushtaqmalik6709 I don't know if SmartPLS does mediated moderation. This video only shows moderated mediation. As for Model 14, I'm not sure what that is. I assume it is something with PROCESS Macro from Hayes.
@@Gaskination I tried my best to figure out how to run this model in AMOS but I couldn't here are some articles that applied this procedure: 1. Investigating the selected consequences of boreout among cabin crew 2. Estimating, Testing, and Comparing Specific Effects in Structural Equation Models: The Phantom Model Approach
I didn't make a video for Gaussian Copula, since it is still somewhat under debate whether it is broadly applicable or too sensitive for all but a few edge conditions.
@@Gaskination Thank you once again for all your endeavours. Could you, if possible, send me the references where this debate is raging? Kind regards, M
@@michaelpaulse1 Here you go. They also cite several useful references. Endogeneity and Gaussian copula: Becker, J.-M./ Proksch, D./ Ringle, C.M.:. Revisiting Gaussian Copulas to Handle Endogenous Regressors, Journal of the Academy of Marketing Science, Volume 50 (2022), pp. 46-66.
Thanks very much Dr . Very clear and simple.
In one of your earlier videos, I think you mentioned why AMOS is better than PLS based SEM software.
Now your recent videos are about Smart PLS.
So, I wonder if you have changed your mind about using AMOS?
Are you going to use SmartPLS from now on instead of AMOS?
Or are you planning to use AMOS and SmartPLS for different purposes?
AMOS is still definitely better for models with all reflective models, because it allows the estimation of model fit. SmartPLS is better for many other things, such as moderation, mediation, and complex things like segmentation (FIMIX and POS). So, I suppose I like AMOS for measurement models and PLS for structural models :) (except the lack of model fit for structural models...)
Thanks for the video Dr. I got different results in AVE, Cronbach's, and P values of the path when using the Process feature, Consistent PLS, and general PLS.
Cronbach's are all over 0.7 when using cPLS and general PLS, while some are below 0.7 when using PROCESS (0.59 is the worst). AVEs are all above 0.5 when using general PLS while some are below 0.5 (the worst is 0.4) when using cPLS. I remember you said if my indicators are all reflective I should choose cPLS, but the P values of the paths in cPLS are too perfect (all 0.000), while when using general PLS they are more "realistic", one is 0.09 while all others are below 0.05...
My model has three controlled variables but no moderation.
I totally do not know which result to trust. Also, because many of my indicators are not normally distributed, I choose PLS instead of CB-SEM. Thank you for helping!
If all factors are reflective, it is probably best to use cPLS. I would not rely in PROCESS for anything except a causal model, which you can also just do with cPLS.
Thank you so much Dr. I have watched almost all of your videos and thank you so much for helping new researchers like me! Great videos and all of them are very clear! Many many thanks and wish you all the best in your life and research!@@Gaskination
Hi Dr. Gaskin
I was looking for videos that discuss control variables in CB-SEM models but I couldn't find any. I have one question regarding my model. Can I assess the model fit without the control variables or must they be included when checking fit? Thanks!
In a structural model, model fit should be assessed with all variables present that will be present when hypothesis testing. In a measurement model, just include all latent variables.
@@Gaskination Thank you so much, I appreciate it! That answers my question perfectly.
@@Gaskination I have four control variables, that are all categorical in nature, each with four categories. As such, I have created dummy variables and included n-1 when running the model (in order to have a reference category). However, when including these control variables, do I group each CV as a single latent variable or do I include each dummy as its own independent variable? I have seen people do both but I can't seem to get a clear answer.
@@TUPhDude I would include them separately, otherwise it is hard to interpret the coefficients.
Thank you for the video! Does this mean that we can now use "PROCESS" for analyzing Control Variables? Also, don't we need to handle the Dummy Variables? Does SmartPLS 4 somehow automatically handle categorical variables? How about Likert-scale variables as controls? Your response would be highly appreciated and helpful to many. Thanks.
1. Process in SmartPLS can include control variables.
2. Dummy variables must still be created for categorical variables.
3. Likert scales can always be used as controls.
@@Gaskination Huge thanks for the prompt response, Sir! Am I wrong in thinking that Likert scale measured construct is sort of categorical (ordinal). And wouldn't that mean it's necessary to create Dummy variable for all Likert-based constructs?
@@juvandi There is a lot of precedent for using Likert scales intact. Some statistical programs, like Mplus, do label ordinal variables as categorical, but they can still be used like a continuous variable. This is not true of nominal variables like job category or marital status, which need to be split into dummy variables. The difference is that with ordinal variables, an increase in the scale value equates to an increase in the construct trait value, whereas with nominal variables, an increase in the scale value does not equate to an increase in the trait value. For example, if with marital status, 1=single, 2=married, 3=divorced, 4=separated, 5=widowed, 6=living together unmarried, then a 5 on the scale is not "more marital status" than a 2. That just doesn't make any sense. Whereas, for a Likert scale for something like job satisfaction, a 5 ("strongly agree") does imply "more job satisfaction" than a 1 on that scale ("strongly disagree"). Hope this helps.
I've written a more complete response here: gaskination.com/forum/discussion/145/do-likert-scales-need-to-be-broken-into-dummy-variables/p1?new=1
@@Gaskination Thank you Dr. Gaskin! Will read the post carefully!
Thank you, Prof. Gaskin. Please what about model fit, Gaussian copulas and NCA?
The SmartPLS website and books strongly discourage the assessment of model fit in PLS models. They have only reluctantly included any model fit features. I didn't make a video for Gaussian Copula, since it is still somewhat under debate whether it is broadly applicable or too sensitive for all but a few edge conditions. I'm not sure what NCA is.
@@Gaskination Thank you for your reply 🙏. NCA = Necessary conditions Analysis
Hello! Hope you are doing great. Actually I need guidance regarding usage of smart pls process feature.
If we select process feature of smart pls 4 for structural model assessment, what we will do for measurement model assessment? Will we combine both the techniques of smart pls 4 i.e. pls sem for assessment of measurement model and process for structural model?? Kindly guide as I couldn't find any suitable answer from literature.
Yes, you can assess the measurement model in CB-SEM (if all reflective and good sample size) or PLS-SEM if it includes small sample size or formative factors.
Dear Dr. Gaskin,
great and helpful video - again!
But I have a question regarding the moderators in PROCESS, and it would be helpful if you could advice me..
In path coefficients, any moderation is not significant.
In conditional direct effects, however, some are significant.
As an example, I would like to test the hypothesis that A (IV) on B(DV) is moderated by age so that its effect is stronger for older people.
If A -> B conditioned on age is significant at +1 SD and its Pathcoefficients are also greater than both, -1SD and at the mean, am I allowed to accept my hypothesis based on the specified conditions? (Even if in path coefficients AxAge→ B is not significant? )
I would be more than grateful if you could help me
Kind regards,
Hannah
There are many approaches to testing moderated effects. The most rigorous is probably floodlight analysis (not available in SmartPLS). The next is probably the simple-slopes analysis (+/- 1 std deviation). The least valid is probably looking directly at the coefficient of the interaction effect.
Thanks Dr. James. Can you throw some light on "Curvilinear Moderation"? Is it possible to conduct in smartPLS or any other software?
I cannot find a way to apply a quadradic effect to a moderation in SmartPLS 4. Perhaps a workaround would be to calculate an interaction term outside of SmartPLS, and then add a quadradic effect to the line that goes from the interaction to the DV.
If we choose pls sem or process sem, is it necessary to run CFA before measurement and structural model assessments?
If you have latent factors, it is important to validate their measurement properties before assessing structural hypotheses.
@@Gaskination these measurement properties should be checked through cfa or measurement model of pls sem?? If we are going to check structural hypothesis through process feature of smart pls 4?
@@nighatyasmeen1 The measurement model of PLS is a CFA. If all factors are reflective, then you can do it through the CB-SEM features of SmartPLS.
@@Gaskination okay thank you
Pardon me Dr. Gaskin, may I know what is differences between conditional direct effects and conditional indirect effects? Could you interpret it? thank you so much :)
The word "conditional" just refers to moderation, and "indirect" refers to mediation. So, a conditional direct effect is just a moderated effect. A conditional indirect effect is the moderated mediated effect.
@@Gaskination thank u for clarifying Dr. Gaskin 😄
Is it true that process is a method to test a theoretical framework in which there is mediation and moderation simultaneously?
Yes, PROCESS can do mediated moderation and moderated mediation.
@@Gaskination Prof can Smart PLS 4 also do mediated moderation and moderated mediation at the same time?
@@mushtaqmalik6709 I don't know if SmartPLS does mediated moderation. This video only shows moderated mediation. As for Model 14, I'm not sure what that is. I assume it is something with PROCESS Macro from Hayes.
@@Gaskination Thanks Prof. These Models relate to PROCESS Macro.
Wonderful explanation. where we can have a practice data set of this video
This dataset, and most others from my videos, can be found on the homepage of statwiki: statwiki.gaskination.com/
Professor, how to get SmartPLS 4 ?
It should be released in June 2022.
Thank you Dr. Very hepful. Can I check if you can do a demo on robustness test using SmartPLS 4？ Thanks =）
I'll have to look into this when I have a moment. Thank you for the suggestion.
Dear Dr.
I have searched a lot to find a video about Phantom variables in SEM, but I couldn't. Would you please give us some information about this?
I do not know what are phantom variables. Perhaps this is related to endogeneity (lurking variables, hidden variables).
@@Gaskination I tried my best to figure out how to run this model in AMOS but I couldn't
here are some articles that applied this procedure:
1. Investigating the selected consequences of boreout among cabin crew
2. Estimating, Testing, and Comparing Specific Effects in Structural Equation
Models: The Phantom Model Approach
@@saeidnosrati3782 I have never seen this approach. I would recommend reaching out to the authors of those papers to see if they can help. Good luck!
Is the conditional effects option available in the WIndows Version of Smart PLS?
The version in this video is on Windows.
Thank you professor. Can you please do a video on SmarPls 4 Gaussian Copula and endogeneity?
I didn't make a video for Gaussian Copula, since it is still somewhat under debate whether it is broadly applicable or too sensitive for all but a few edge conditions.
@@Gaskination Thank you once again for all your endeavours. Could you, if possible, send me the references where this debate is raging? Kind regards, M
@@michaelpaulse1 Here you go. They also cite several useful references.
Endogeneity and Gaussian copula: Becker, J.-M./ Proksch, D./ Ringle, C.M.:. Revisiting Gaussian Copulas to Handle Endogenous Regressors, Journal of the Academy of Marketing Science, Volume 50 (2022), pp. 46-66.